eAppendix. Supplementary Methods
eTable 1. Alternate Specifications
eTable 2. Alternate Specifications – Include Out of Network Claims
eTable 3. Complete Regression Results From Main Models
eTable 4. Complete Results From Models of Changes Over Time
Baker LC, Bundorf MK, Royalty AB, Levin Z. Physician Practice Competition and Prices Paid by Private Insurers for Office Visits. JAMA. 2014;312(16):1653-1662. doi:10.1001/jama.2014.10921
Physician practice consolidation could promote higher-quality care but may also create greater economic market power that could lead to higher prices for physician services.
To assess the relationship between physician competition and prices paid by private preferred provider organizations (PPOs) for 10 types of office visits in 10 prominent specialties.
Design and Setting
Retrospective study in 1058 US counties in urbanized areas, representing all 50 states, examining the relationship between measured physician competition and prices paid for office visits in 2010 and the relationship between changes in competition and prices between 2003 and 2010, using regression analysis to control for possible confounding factors.
Variation in the mean Hirschman-Herfindahl Index (HHI) of physician practices within a county by specialty (HHIs range from 0, representing maximally competitive markets, to 10 000 in markets served by a single [monopoly] practice).
Main Outcomes and Measures
Mean price paid by county to physicians in each specialty by private PPOs for intermediate office visits with established patients (Current Procedural Terminology [CPT] code 99213) and a price index measuring the county-weighted mean price for 10 types of office visits with new and established patients (CPT codes 99201-99205, 99211-99215) relative to national mean prices.
In 2010, across all specialties studied, HHIs were 3 to 4 times higher in the 90th-percentile county than the 10th-percentile county (eg, for family practice: 10th percentile HHI = 1023 and 90th percentile HHI = 3629). Depending on specialty, mean price for a CPT code 99213 visit was between $70 and $75. After adjustment for potential confounders, depending on specialty, prices at the 90th-percentile HHI were between $5.85 (orthopedics; 95% CI, $3.46-$8.24) and $11.67 (internal medicine; 95% CI, $9.13-$14.21) higher than at the 10th percentile. Including all types of office visits, price indexes at the 90th-percentile HHI were 8.3% (orthopedics; 95% CI, 5.0%-11.6%) to 16.1% (internal medicine; 95% CI, 12.8%-19.5%) higher. Between 2003 and 2010, there were larger price increases in areas that were less competitive in 2002 than in initially more competitive areas.
Conclusions and Relevance
More competition among physicians is related to lower prices paid by private PPOs for office visits. These results may inform work on policies that influence practice competition.
Physicians are increasingly moving away from solo and smaller practices toward larger organizations.1- 3 These changes may be beneficial if larger practices with more resources are better able to coordinate care, adopt process improvements, increase use of information technology, or take other actions that improve quality of care.4- 7 At the same time, a trend toward fewer and larger groups could increase what economists refer to as “market concentration,” resulting in fewer practices facing less competition and with greater economic power. This in turn could lead health plans to pay higher prices for physician services.8,9
A number of studies document that hospitals and health insurance companies with more market power do receive higher prices for their services.8 However, there is little evidence on the relationship between competition and prices paid for physician services, with studies limited to cardiology and orthopedics10 and physicians in California.11
Additional evidence on the relationship between physician practice competition and prices paid for physician services would shed light on how recent changes in physician organization have affected health care spending and help guide policy toward competition in health care markets.12 It may also help identify the sources of the wide variations observed across areas in prices paid to physicians for similar services.13- 15 We performed a study to assess the relationship between physician competition and prices paid by private preferred provider organizations (PPOs) for 10 types of office visits to physicians in 10 prominent specialties.
We used data on prices paid to physicians from the Truven Analytics MarketScan Commercial Claims and Encounters database for 2010.16 This database contains information from adjudicated and paid claims filed for the care of more than 49 million privately insured individuals with employment-based insurance through a participating employer. Although the database disproportionately draws from the South, it includes claims from many insurers, has very wide overall geographic coverage, and has been frequently used in analyses of health care payments and spending.15,17- 20
We studied prices paid for office visits with new patients (Current Procedural Terminology [CPT] codes 99201-99205) and established patients (CPT codes 99211-99215). These 10 evaluation and management codes are among the most commonly billed in the United States and accounted for 25% of claims and 23% of payments by PPOs in the database. Codes 99201-99205 include all office visits for new patients not previously seen by the same physician. The 5 included services are intended to span the range of intensity across office visits from code 99201, a “basic” office visit with minimal patient complexity and a short duration, to code 99205, an “advanced” office visit with a high degree of complexity and a long duration. The range 99211-99215 similarly captures the range of office visits for established patients, who have been previously seen by the physician. We included only payments from health plans identified as PPOs that paid physicians on a fee-for-service basis. We studied the amount the plan agreed to pay the physician for the service after the application of contractual discount provisions and other plan rules, commonly called the “allowed amount.” We refer to this as the “price” for the service. The physician may have received this partly from the insurer and partly from the patient in the form of applicable co-payments or deductibles.
We compiled prices paid to physicians in 10 large, clinically diverse specialties that generate large numbers of office visit claims: internal medicine, family practice, cardiology, dermatology, gastroenterology, neurology, general surgery, orthopedics, urology, and otolaryngology. In addition to internal medicine and family practice, this set includes the 4 largest medical and the 4 largest surgical subspecialties in the database. We were unable to include pediatrics and obstetrics/gynecology because our practice competition measures are derived from Medicare data, in which these specialties are incompletely represented.
For each CPT code, for each US county, for each specialty, we obtained the number of claims and mean price paid to physicians reporting a practice location in the county. Claims were included if the patient was younger than 65 years, the provider of the billed service was identified as an in-network physician, the reported place of service was a physician office, and the claim was for professional services (as opposed to facility charges). A small number of claims with prices more than 100 times or less than 0.01 times the national mean for the given CPT code were excluded as outliers.
For some analyses, we aggregated the 10 CPT codes into a specialty-specific county price index for office visits. The index is the amount paid in the county for the 10 CPT codes divided by the amount that would have been paid had the claims from county physicians been paid at the specialty-specific national mean price for each CPT code. An index above 1 indicates a county where mean office visit prices paid exceeded the national mean, and vice versa (see the eAppendix in the Supplement for further detail).
Physician negotiations with health plans over prices paid in a given year usually take place in the year prior, if not earlier; therefore, we linked the price measures to practice competition measures from the preceding year. Because the MarketScan data do not contain enough information to measure competition, we derived our competition measures from Medicare claims filed by physicians for the care of a 20% random sample of traditional Medicare enrollees. Medicare claims reflect care delivered by a very large share of active physicians, and the set of physicians who billed traditional Medicare should substantially overlap with the set of physicians who could have provided services to private PPO patients. Each Medicare claim reports the tax identification number (tax ID) of the physician’s practice, the physician’s specialty, the physician’s practice zip code, and the patient’s residence zip code.
Following previous work,3,21- 25 we identified a practice as a group of physicians reporting the same specialty who billed under the same tax ID. Physicians who use the same tax ID are part of the same financially integrated organization. Many financially integrated organizations use the same tax ID for all physicians in their organization, but it is permissible for the same organization to use multiple tax IDs. Because physicians practicing together in financially integrated organizations may legally bargain jointly over fee-for-service prices,26 this is an appropriate practice definition for our study. Other non–financially integrated organizations, like independent practice associations, which do not use common tax IDs, would not be identified by this approach, but these types of organizations may not normally bargain jointly over fee-for-service payments.27
We used a standard economic competition measure, the Hirschman-Herfindahl Index (HHI).28 The HHI is the sum of the squared market shares of practices serving a market multiplied by 10 000. Markets served by a large number of practices, each with a small market share, will have a low HHI, signaling a more competitive market. For example, a market with 10 practices, each of equal size, would have an HHI of 1000. The HHI increases, signaling less competition, when there are fewer practices or when the market share of any one practice increases relative to the others. For example, a market with 4 practices with equal market shares of 25% would have an HHI of (0.252 + 0.252 + 0.252 + 0.252) × 10 000 = 2500. A market with 4 practices, one with 70% share and the others with 10%, would have an HHI of (0.102 + 0.102 + 0.102 + 0.702) × 10 000 = 5200. The HHI reaches its maximum of 10 000 in a monopoly market, served by a single practice.
The HHI is widely used in analyses of competition.29 An attractive feature of the HHI is that it has proven useful for analyzing many industries of differing sizes and is suitable for studying physicians in different specialties and markets. The HHI is a primary tool used by the Federal Trade Commission (FTC) and Department of Justice (DOJ) to evaluate mergers and assess the public policy implications of changes in competition.27,28
Adapting methods developed for hospitals30 and incorporating the FTC and DOJ recommendations for measuring competition for accountable care organizations,23 we constructed a specialty-specific HHI for each practice. We first identified the zip codes from which the practice drew patients. We then calculated an HHI for each zip code based on the market shares of practices serving patients in the zip code and constructed a practice-level HHI as the mean of the zip code HHIs for each zip code in the practice’s market. To match our county-level price data, we constructed county-level means of the HHIs of the practices of physicians located in each county (see the eAppendix in the Supplement for further detail and a discussion of validation exercises).
We restricted our analysis to counties within Core Based Statistical Areas, defined by the Census Bureau as groups of counties tied to urban centers of 10 000 people or more. We further excluded county-specialty combinations with fewer than 5 underlying claims for CPT code 99213 office visits.
We present results for 2 price measures: prices paid for intermediate office visits with established patients (CPT code 99213), the most commonly billed of the office visits we studied, and the office visit price index. We used ordinary least squares linear regression to examine the association between HHI and price measures. The main independent variable was the HHI, measured continuously. We estimated 3 versions of this model, varying the set of included control variables. In the first, we included no controls to observe the unadjusted relationship between prices and HHI. In the second, we included a set of controls designed to adjust for characteristics of counties that could influence prices, including county population, total number of physicians per population, number of physicians in the given specialty per population, number of short-term general hospitals and hospital beds per population, median household income, percentage of the population uninsured, percentage of the population older than 25 years who completed high school, percentage of the population older than 25 years who completed 4 or more years of college, percentage of the population enrolled in Medicare, and percentage eligible for Medicaid. We used the Medicare geographic practice cost indexes to control for practice costs.
Because the regression models controlled for the number of physicians per capita, the associations between HHI and prices we measure should be interpreted as reflecting changes in the ways physician practices are organized, statistically holding fixed the number of physicians. That is, they may be interpreted as showing, for a given number of physicians, how prices vary when those physicians are organized into larger rather than smaller practices.
In the third model, which adopted the most conservative approach, we included all of these variables as well as state fixed effects. The fixed effects capture characteristics of states that we did not observe but could have been correlated with competition and prices, though at the risk of causing us to underestimate the true strength of the association between HHI and prices.
We estimated the models separately for each of the 10 specialties to allow for variation in the association between competition and prices across specialties and report the estimated change in the price measure associated with a 1000-point increase in the HHI. We also present the implied difference in prices associated with being in the 90th-percentile HHI county compared with the 10th-percentile county, separately for each specialty.
eTables 1 and 2 in the Supplement report results from sensitivity analyses, including considering nonlinear models of HHI, controlling for the presence of multispecialty groups, including a measure of the HHI of area PPOs,31 and including claims for out-of-network services, all of which produced results consistent with those reported.
Studying the relationship between changes in HHIs and changes in prices within counties over time could help control for unobservable county-level confounders. We constructed measures of 2003 payments and 2002 HHIs analogous to those described above. The 2003 price indexes measure 2003 county prices relative to 2010 national mean prices so that changes in the price index will measure growth in prices over time. Within each specialty, we selected counties with at least 5 claims for CPT code 99213 visits in both 2003 and 2010. To allow separate examination of associations between competition changes and price changes for areas that were initially more and less competitive, we divided counties using their 2002 HHI into initially more competitive (2002 HHI <2500) and less competitive (2002 HHI ≥2500) groups, based on the FTC and DOJ definition of “highly concentrated” markets.27 We estimated an ordinary least squares regression, by specialty, in which the dependent variable was the 2003-2010 price measure change. The independent variables included an indicator for areas that were initially less competitive to measure differences in price change associated with the initial HHI. The independent variables also included a linear continuous measure of the change in the HHI interacted with the initial HHI level to measure the association between changes in HHI and changes in price separately for areas that were initially more and less competitive. The models also controlled for changes in the time-varying characteristics described above.
Throughout the analysis, we computed robust standard errors to account for variation in the number of claims underlying the dependent variables. Evaluations of statistical hypotheses were conducted using 2-sided tests. We considered results with P≤.05 to be statistically significant. Statistical analysis was conducted using SAS version 9.3 (SAS Institute Inc) and Stata version 13 (Stata Corp). The Stanford University and Indiana University institutional review boards approved the study protocol and granted a waiver of consent.
A county-level observation was available for at least 1 specialty in 1058 of 1098 total US counties in Core Based Statistical Areas, representing all 50 states and the District of Columbia and 85% of the total US population in 2010. Many counties had a sufficient number of claims to support analysis for some but not all specialties, so the number of counties used in analysis varied across specialties from 611 in neurology to 1052 in family practice.
The level of competition varied across specialties (Table 1). Family practice and internal medicine practices faced the most competition (mean practice HHIs of 2139 and 1744, respectively). More specialized fields tended to have higher HHIs, indicating less competition. There was considerable variation across counties within specialties. The mean practice HHI in the 90th-percentile county was typically 3 times higher than in the 10th-percentile county. Many counties had HHIs of 2500 or higher, the level at which the FTC and DOJ classify a market as highly concentrated.28
Mean prices for CPT code 99213 visits were similar across the 10 specialties, consistent with other work.15 Within each specialty, prices varied noticeably across counties. The ratio of the 90th to 10th percentiles varied between 1.51 (family practice) and 1.68 (neurology). The office visit price index showed similar variation. Within specialties, the price index showed that 10th-percentile counties typically had prices 18% (family practice) to 23% (neurology) lower than the national mean, while the 90th-percentile counties had prices 23% (family practice) to 29% (neurology and otolaryngology) higher.
Characteristics of counties with HHIs higher than the specialty median HHI and counties at or lower than the median are shown in Table 2 for 3 representative specialties. Counties where the mean practice HHI was higher than the median had smaller populations than areas with lower HHIs and tended to have more physicians and hospitals per capita, perhaps owing to their smaller population. Higher-HHI counties also tended to have moderately lower income and educational attainment, uninsurance rates, and geographic practice cost index levels.
In unadjusted regression analysis and after adjusting for county characteristics, mean payments for office visits were lower in areas with more competition (Table 3). Complete results of the regression analyses are available in eTable 3 in the Supplement. After adjustment, a 1000-point-higher HHI was associated with statistically significantly higher CPT code 99213 visit prices in all 10 specialties, ranging from $1.41 (urology; 95% CI, $0.70-$2.12) to $4.69 (internal medicine, 95% CI, $3.76-$5.62). The results for the price index are similar. After adjustment, a 1000-point-higher HHI was associated with values of the index between 0.020 (urology; 95% CI, 0.010-0.029) and 0.065 (internal medicine; 95% CI, 0.052-0.077) higher. This can be interpreted as indicating prices between 2.0% and 6.5% higher, relative to national mean prices.
Across these 10 specialties, the 90th-percentile county’s HHI was approximately 2500 to 4500 points higher than the 10th-percentile county’s HHI. Based on the estimates shown in Table 3 and the HHI data shown in Table 1, overall mean office visit prices were between 8.3% (orthopedics; 95% CI, 5.0%-11.6%) and 16.1% (internal medicine; 95% CI, 12.8%-19.5%) higher in the 90th-percentile HHI county relative to the 10th-percentile county.
In the third regression model (Table 4), inclusion of state fixed effects reduced the magnitude of the association in all specialties, but a statistically significant association was still observed in 8 of the 10 specialties for both the CPT code 99213 price and the price index.
The preceding analyses focused on cross-sectional variation in 2010. Analyses of 2003-2010 changes in prices paid are reported in Table 5 for counties for which sufficient data were available in both 2003 and 2010. Complete results of the regression analyses are available in eTable 4 in the Supplement. Two principal patterns are evident. First, prices increased more in areas that were initially less competitive than in those that were more competitive. For example, for CPT code 99213 visits for internal medicine, the change in prices between 2003 and 2010 was $4.64 greater for initially less competitive areas (95% CI, $2.49-$6.80; P < .001). Second, in some specialties, areas that were initially more competitive had increases in price when they became less competitive, but areas that were initially less competitive did not. For example, for CPT code 99213 visits for internal medicine, a 1000-point increase in HHI between 2003 and 2010 in an area that was initially more competitive was associated with a $2.33 increase in price (95% CI, $0.98-$3.69; P < .001), but there was no significant association in initially high-HHI areas.
Less competition among physician practices is statistically significantly associated with substantially higher prices paid by private PPOs to physicians in 10 large specialties for office visits. Across the 10 specialties we studied, estimates from our main model imply that the level of competition observed at the 90th percentile of the HHI distribution was associated with a price for an intermediate office visit with an established patient (CPT code 99213) between $5.85 and $11.67 higher than at the 10th percentile of the HHI distribution. Across all 10 types of office visits, this difference in HHI was associated with mean prices for office visits 8.3% to 16.1% higher. In our more conservative model, this difference in the HHI was associated with 3.5% to 5.4% higher mean prices. This is consistent with the hypothesis that greater market power allows physicians to bargain for higher prices from private insurance companies.
Examining changes in prices between 2003 and 2010, we found that prices increased more rapidly in areas where practices were initially less competitive than in other areas. In some specialties, declining competition was also associated with larger increases in prices in areas that were initially more competitive. This pattern suggests the possibility that the results we observe in 2010 may be related to the ability of practices in low-competition areas to negotiate larger price increases over time as well as related to changes in competition over time. This suggests that a lack of competition could have long-lasting effects, continuing to drive future price increases even without further changes in HHI.
An association between competition and prices may have important implications for health policy, as pressures to increase practice size persist or even increase in the future.12,32 We saw substantial amounts of concentration in the markets we studied, which raises concerns about potentially harmful implications for consumers.28 Higher health care spending due to increased prices paid to physicians without accompanying improvements in quality, satisfaction, or outcomes would generate inefficiency in the health care system. The United States spent nearly $250 billion on privately insured physician services in 2011,33 so even small percentage increases could result in tens of billions of dollars in additional spending.
We present results from models that included state fixed effects and models that did not. State fixed effects statistically control for all baseline characteristics of states. However, while fixed effects may helpfully control for potential confounders, they are also a blunt instrument that may “overcontrol,” resulting in estimates of the relationship between competition and prices that are too conservative. This approach does not allow the estimated association between competition and prices to be influenced by differences in competition across states. If variation in competition across states were an important driver of prices paid, estimates that excluded this variation from consideration could easily understate the true relationship.
Fully interpreting the implications of these results for the overall performance of the health care system requires additional information. If larger organizations systematically produce higher quality, higher prices may be justified. Larger organizations may also offer other advantages such as ease of contracting with health plans or the ability to adapt to evolving financial incentives. Even so, the association between higher prices and declining competition clearly points to the importance of gathering information and monitoring ongoing changes to mitigate potential harm to consumers.
The results of this study should be interpreted in light of several limitations. First, our primary results are derived from cross-sectional analyses, which precludes any determination of causality. It is possible that our results are confounded by characteristics of areas that are associated with both competition and prices paid but that we did not observe. We did take steps to control for a number of observable factors that could have confounded the results, but these are necessarily incomplete. A potentially important but difficult to observe confounder is the degree of competition between private insurers. Although available measures of PPO competition have significant flaws,34 we performed sensitivity tests using one such measure and found that it did not significantly affect our conclusions (eTable 1 in the Supplement). Furthermore, to attempt to control for insurer competition and any other unobserved but relevant characteristics of states, we estimated models that included state fixed effects. That we found consistent results from these approaches increases our confidence that our main results are not due to unobserved differences across areas.
Second, potential exists for reverse causality. It may be that prices influence the propensity to form larger physician organizations. We cannot rule this out, but we reasoned that physicians would be more likely to consolidate in response to low prices than to high prices, so the likely effect on our results would be to understate the true strength of the relationship between HHIs and prices.
Third, we studied prices paid by PPOs, which may not be representative of other types of plans. Our PPO data are drawn from the MarketScan database, a convenience sample of employer-provided insurance plans which, while large and geographically diverse, may not be representative of all private PPOs.
Fourth, our results rely on measures of competition derived using tax IDs to identify practices. There is reason to believe that this approach characterizes variations in practice structures with sufficient accuracy to generate useful results. However, practices can be complex entities that evolve over time, and any indicator of practice structure derived from broad-based data sources is bound to be imprecise for at least some practices. Using tax IDs may cause us to overestimate the amount of competition in some situations. When practices consolidate, there may be circumstances in which they prefer to adopt a single unified tax ID and other circumstances in which they prefer to retain separate tax IDs. When hospitals acquire practices, in some cases all acquired practices adopt the tax ID of the acquiring institution, but not always. In these cases, retaining separate tax IDs would cause us to treat practices as competitors when in fact they may not be. We cannot rule out the possibility that misestimation of the level of competition has affected our estimates, and we cannot determine the direction of any resulting bias, but we were able to conduct some validation exercises and robustness checks that are consistent with the usefulness of the results we present.
Study findings show that less competition between physicians is related to higher prices paid to physicians by private PPOs for common office visits. These results may inform the development or adaptation of policies that influence practice competition.
Corresponding Author: Laurence C. Baker, PhD, Department of Health Research and Policy, Stanford University School of Medicine, Stanford, CA 94305-5405 (firstname.lastname@example.org).
Author Contributions: Dr Baker had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Baker, Bundorf, Royalty.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Baker.
Critical revision of the manuscript for important intellectual content: Bundorf, Royalty, Levin.
Statistical analysis: All authors.
Obtained funding: Baker, Bundorf.
Administrative, technical, or material support: Baker.
Study supervision: Baker.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Drs Baker, Bundorf, and Royalty report receiving consulting fees from the National Institute for Health Care Management during the conduct of this study for participation in an award selection screening panel. Dr Baker reports receiving consulting fees from Kaiser Permanente and the American Hospital Association. No other disclosures were reported.
Funding/Support: The National Institute for Health Care Management provided funding to support this work.
Role of the Funders/Sponsors: The funder had no role in the design or conduct of the study or the collection, analysis, or interpretation of data; in the preparation, review, or approval of the manuscript; or in the decision to submit the manuscript for publication.